Vascular Dynamics Aid a Coupled Neurovascular Network Learn Sparse Independent Features: A Computational Model

Area of Science:

• Neuroscience
• Computational Biology
• Systems Biology

Background:

• Cerebral vascular dynamics are traditionally viewed as solely controlled by neural activity.
• Emerging evidence suggests a feedback loop where vascular dynamics modulate neural function.
• The hemoneural hypothesis posits that vascular feedback, via glucose and oxygen, influences neuronal firing.

Purpose of the Study:

• To investigate the functional consequences of vascular feedback on neural network dynamics.
• To model the interplay between a vascular network and a neural network.
• To explore how vascular dynamics impact information processing in an auto-encoder.

Main Methods:

• Developed a computational model of coupled vascular and neural networks.
• Modeled vascular units as oscillators supplying energy to hidden neurons in an auto-encoder.
• Linked vascular network dynamics to the auto-encoder's reconstruction error, representing neuronal demand.

Main Results:

• Desynchronized vascular dynamics led to reduced auto-encoder reconstruction error.
• Learned feature vectors were sparse and independent under desynchronized vascular conditions.
• Demonstrated that desynchronized vascular dynamics promote efficient training of the auto-encoder network.

Conclusions:

• Vascular feedback significantly modulates neural network activity and information processing.
• Desynchronized vascular dynamics are crucial for efficient neural network training and learning.
• This study supports the hemoneural hypothesis by providing a computational framework for vascular-neural interactions.